Justin Mason wrote:
On Tue, Mar 3, 2009 at 17:40, Marc Perkel <m...@perkel.com> wrote:
John Wilcock wrote:
I wonder about the feasibility of a second Bayesian database, using the
same learning mechanism as the current system, but keeping track of rule
combinations instead of keywords.
YES! That is something I think is worth trying.
this is worth investigating, I know someone was considering trying it
in the past. A key aspect apparently is to make a token representing
the *lack* of a rule hit -- ie. the rule was run, but did not fire.
--j.
I agree. Make a "rule passed" token as well and feed that data into some
sort of statistical engine to see if we can automatically identify
patterns that are usable.